Poster abstract: Distributed coordination of sub-systems power-modes and software-modes

Summary form given. Energy management is essential for cyber-physical systems. Such systems typically consist of several, often distributed, sub-systems that may communicate. State-of-the-art hardware blocks employed in these sub-systems have several power-modes that can be controlled to consume less energy. To-date, the decision of power-modes is most of the times taken within each of the sub-systems. However, it does not consider neither the external, general context of the system, nor the software-modes which involves on the Quality of Service (QoS) of the system. This may lead to large energy waste. To address this problem, we propose a loosely coupled and distributed framework that selects the appropriate sub-system power-mode. The selection takes into account both external context (e.g. GPS location, ambient temperature, information from external applications) that cannot be directly accessed on a sub-system, and software-modes. The flexibility of the framework allows to control, at the same time, the power-modes of sub-systems and the QoS of the system using the same primitives. The framework is based on the LINC coordination middleware [1] which has several interesting properties. LINC handles synchronization between subsystems by grouping a set of operations into transactions that provide an all-or-nothing property. This ensures, for instance, that the external information which triggers a power-mode switch, is still valid when the power-mode is actually updated. Furthermore, LINC hides the heterogeneity of the controlled devices or communication protocols by using an abstraction layer based on associative memory. LINC also offers a high level programming model that allows to describe a sequence of operations as goal-driven rules. The proposed framework is evaluated in a vehicle system that includes among other a GPS, speed sensors, and the STHORM platform [3] which is a state-of-the-art many-core system-on-chip. A perception application, based on the prin- iple of the Bayesian Occupancy Filter (BOF) [2], runs on the platform and detects obstacles in the environment around the vehicle. The BOF discretizes the environment into a grid whose resolution represents the QoS of the system. The resolution is selected according to the speed and location of the vehicle (e.g. in a city or on highway). Each value of the resolution has different processing requirements which allow to scale the power-mode depending on the vehicle location and speed. For instance, on highway, the vehicle runs at a high speed, and in a less diversified environment than in city. Then, a lower resolution can fit the application. This requires less computation power, and consequently, a lower power-mode can be selected. The results have shown that we can realize a significant power-saving by considering external context and software-modes when selecting sub-systems powermode. The loosely coupled approach of LINC eases the integration of any cyber-physical systems into the framework. In addition, the goal driven production rules simplify the coordination of the whole system.